a 850 Environ. Sci. Technol.. Val. 26,No. 5. 1992
d
0013-936X192/0926 850$03 0010 0 1992 American Chemical Society
€SmFEATURES
MODELING THE I N D O O R
ENVIRONMENT In this paper we briefly review currently available computer models that estimate human exposure to indoor pollutants. We examine representative indoor air models and their input data requirements from the standpoint of their relative effectiveness in a variety of problem conditions. This paper is not a definitive, systematic evaluation of the currently available databases and models. Rather, it aims to help the model user [decision maker) develop a process by which to evaluate the strengths, limitations, and range of applicability of available models and data. Understanding the inherent accuracy of our analytical tools is particularly relevant when these tools are used to assess health risks. When they know the quantitative uncertainty of the analytical results, decision makers can select the most appropriate risk-reduction strategy and can specify, within established confidence limits, the level of each control measure needed to achieve a desired risk reduction. When the magnitude of risk uncertainty does not permit one to distinguish between solutions with significantly different potential social and economic impacts, one cannot simply increase the degree of some single control to accommodate the inaccuracy of the analytical approach utilized. The analyst must understand the relative effectiveness of the tools available and the accuracy with which they can be applied to a specific problem. The decision maker can then select the optimal analysis of the prob-
Barbara S. Austin Stanley M. Greenfield Bruce R.Weir Gerald E. Anderson Systems Applicotions Znternational Son Rafael, CA 94903 Joseph V. Behar U.S. Environmental Protection Agency Office of Research b Development Las Vegas, N v 89193 lem posed, as well as acquire more information that will permit it to be used effectively. The past two decades have witnessed the development of increasingly sophisticated mathematical models for describing ambient [outdoor) air quality. This was in response to the demand for more effective emission control regulations to reduce environmental risk, and the resulting need for tools with which to analyze the complex ambient environment. It is recognized that the simplistic air quality models developed early in this period do not adequately represent the total ambient environment. More re-
cently, concern for reducing environmental risks has focused on the need to consider the total exposure of a population-indoors as well as outdoors. Considerable ambient and personal monitoring data are now available that more accurately reflect the complex variation in time and space of the indoor and outdoor environments. These data indicate that both outdoor and indoor air pollutant concentrations contribute to health risk for most people. Indoor exposures result from the temporal and spatial variations of indoor sources and sinks, chemical transformations, population dynamics, and background concentrations of various pollutants that are directly or indirectly related to outdoor ambient air quality. Indoor temporal and spatial variations are so complex that the simple averages of these variables given by earlier models may not adequately express indoor exposures. The need to portray the indoor environment more accurately will lead to better databases and models. The ability to assess model capabilities and database accuracy will, ultimately, determine the effectiveness of the analytical tools we create. Estimates of the human health risk from toxins in the indoor or outdoor air require estimation of patterns of air concentrations, the degree of transfer to other exposure media (e.g., the deposition of aerosols onto surfaces from which they could be ingested through the skin or mouth), the dose-receiving activEnviron. Sci. Technol.. VoI. 26, No. 5, 1992 051
ity of human “receptors” in the contaminated areas, and the doseresponse relationships for the specific toxins present. Each of these elements has been addressed with both monitoring and modeling techniques, mostly for outdoor exposures, although some techniques have been applied to indoor problems. For example, in 1978 EPA began developing the Human Exposure Model (HEM) in support of its hazardous pollutant program ( 1 ) . In 1980 the Agency began developing t h e N a t i o n a l E x p o s u r e Model (NEM) ( z ) , which addressed both indoor a n d outdoor exposures. HEM was based purely on longterm (lifetime) dispersion modeling. NEM, although capable of using model-defined patterns, was initially based on the use of outdoor monitoring and empirically based indoor-outdoor relationships ( 3 ) . Indoor exposure estimates used in models such as NEM are limited to global estimates of indooroutdoor relationships; in this context, uncertainty may be low, but the information content in the resulting exposure estimates also may be low. Many sophisticated models of indoor airflow patterns were developed and used by the building design industry, but i n general these models do not address concentration patterns caused by indoor emissions; thus, uncertainty is not a relevant issue in the health risk context. Some indoor models address emission processes in significant detail; however, because these models seldom address more than the mean zonal concentration, and few address human activity patterns, uncertainty means little pragmatically in this case. As in the case of ambient models, however, users will constantly demand more sophisticated models and an ability to represent pollutant concentrations in the complex indoor environment with higher temporal and spatial resolutions. If these higher resolutions are used to “better” determine the potential health risks involved, then the uncertainty of their determination will affect considerably the credibility of the assessment. In the next sections we describe several indoor and exposure models we consider representative of the many that have been developed for specific conditions, and some considerations associated with population activity data used in exposure models. 852 Environ. Sci. Technol., Vol. 26, No. 5,1992
Indoor and exposure models
Many regulatory issues require the use of risk assessments containing human exposure estimates. The significant contribution of indoor air quality to an individual’s total exposure (most people spend well over 90% of their time indoors) requires a more systematic approach for choosing how to estimate exposure than is currently available. A number of modeling tools can be used to estimate indoor pollutant concentrations and human exposure. Although each model has its strengths and weaknesses and in most cases is continually being improved, few risk assessors are aware that most available models were designed for a specific set of conditions and may not be the best model for their particular purpose. This potential problem can result in errors ranging from gross to subtle. This section presents an initial step in choosing the most appropriate model for a given application: to examine a general set of relevant parameters. A key to choosing the best model for an application is understanding how it addresses various parameters relevant to the application. Table l presents a set of general par a m e t e r s a n d s u m m ar i z e s h 0 w several representative indoor and exposure models address each parameter. Most of our discussion focuses on the models described in the table. We have excluded other equally useful models simply for brevity. An obvious but often overlooked first step is to identify whether the output is consistent with the application, and, if not, whether it can be converted to the desired type without undue effort, increased uncertainty, or loss of precision. For example, if an analyst is studying the range of concentrations of combustion-related pollutants that occur indoors, then MACROMODEL, INDOOR, IAQM, or CONTAM appear to be appropriate; if VOCs are of interest, then MACROMODEL may not be appropriate. If the model output is to be linked with an exposure model, it should consist of a time series of concentrations, with the time steps consistent with the averaging time of interest for the exposure model. T h e way i n w h i c h p o l l u t a n t sources and sinks are characterized clearly is important. In general, the factors that need to be considered for these parameters are as follows:
the pollutants that are modeled, whether there is a focus on particular sources, the time steps that are allowed, whether the model allows the user to specify emission rates that change with time, the amount of detail provided in the model with respect to default values, and the requirements for user input. Pollutant “sink” factors by which indoor pollutants escape (via deposition on room surfaces, reaction with other compounds present in the air, or dispersion to other rooms or outdoors) are of great importance to indoor modeling, yet are rarely treated with much detail. For example, of the models described in Table 1, only IAQM and CONTAM consider the relative speed with which a compound might react with surfaces in the room(s) being modeled. Virtually all indoor models use the surface area of the walls as the key determinant of deposition rates; however, the surface area and composition of the room contents are also relevant (e.g., the effect produced by metal filing cabinets is different from that of pile carpets or tweed sofas). The more sophisticated models provide ways of addressing such issues but require more knowledge on the part of the user. For example, the IAQM ( 5 )is one of the most flexible models presented in Table 1 with respect to its ability to model almost any pollutant from any source for any averaging time. However, the user needs to enter the emission rate for each averaging period as well as a “reactivity factor” (a rate constant associated with heterogeneous surface decomposition used as a pollutant sink). These inputs require that the user be very familiar with studies of indoor emiss i o n rates a n d t h e i r p o t e n t i a l shortcomings. Without this familiarity, the model results can be misleading. The uncertainties introduced by the model formulation, in the inputs themselves, or simply by gaps in the model operations must be known by the user of the estimates. Further, the choice among computational accuracy, flexibility, and user level of effort should be explicit because this choice strongly affects the comprehensiveness of a study’s conclusions. The model CONTAM ( 7 ) and others i n the CONTAM “family” such as DTAM and AIRNET require even more sophistication on the part of the user,
but the rewards are increased accuracy and precision of results. A different approach is embodied in the EPA models INDOOR ( 4 ) and IAQPC (15);they offer default values for several specific sources, such as floor wax, moth crystals, kerosene heaters, and cigarettes, which can be changed by the user if desired. There is also an “other” category for different sources. This approach allows a less experienced user to model with values that are at least reasonable for the abovementioned sources. A model structured in this way cannot address as many conditions as others, but it can be very useful if the sources and pollutants it covers are the ones in which the risk assessor is interested. The way in which models address airflow and building parameters is also crucial, although less experienced users are often not aware of the sensitivity of indoor pollutant concentrations to these factors. Two types of parameters are very significant: airflow parameters, such as the number of air changes per hour (ach) (i.e, how many times per hour the room air is completely exchanged with outdoor air) a n d building parameters that control airflow characteristics-such as the number of doors and windows, the type of HVAC (heating, ventilation, and air conditioning) system, and the size of the rooms. For example, if a room or building were of infinite size, any emission rate would eventually result in essentially zero concentration. Conversely, if a room had nearly zero ach, even a very low emission rate would ultimately result in high concentrations. Before a model is chosen for an application, the way in which these parameters are treated and their ability to represent conditions being modeled must be evaluated. For example, residences that are not equipped with air conditioners may have wide-open windows and doors during the summer. The resulting ach (generally greater than 10) is far higher than that in a house with closed windows and doors and an operating HVAC system (generally around 1 to 3 ach). A model not equipped to handle varying building conditions is not likely to be appropriate for application to summer or annual average conditions. When a model’s treatment of airflow and building parameters is evaluated, issues such as the degree of generalization to the building stock of the geographic
area being studied can be very important. If a risk assessor needs estimates of exposure or concentrations in buildings in an entire geographical region where these parameters vary widely, he or she must be prepared to make separate model runs for each building type in the modeled area (if using a model such as INDOOR) or to accept the definition and distribution of types internal to a specific model. Users of exposure models should be aware of a number of analytical difficulties. For example, an exposure estimate for a population group, such as residents of New York or all asthmatics in the United
The significant contribution of indoor air quality to an individual’s total exposure (most
people spend well Over
90% of their time indoors) requires a systematic approach for choosing how to estimate exposure. States, needs to address the wide variation in indoor environments in which individuals spend time during the day, each of which has different concentrations. Concentration estimates for each of these environments, as well as for the environment an individual is in at a given time, can be very uncertain. For example, a concentration model may predict an unhealthy concentration of a compound in a kitchen at 6:OO p.m. How likely is it that a particular i n d i v i d u a l is i n the kitchen a t that particular time rather than still at work or in another room in the same home? In addressing these questions,
most exposure models fall into one of three categories: “what i f ’ models such as EXPOSURE (16) in which a user makes the assumption that a person “is in the kitchen at 6:OO p.m,” then enters a specified activity pattern to estimate exposure; “deterministic” models such as PAQM or NEM, which group the population into numerous subgroups, each of which has exactly the same hourly pattern of activities: and “probabilistic” models such as SHAPE or BEAM, which treat population movement stochastically. The stochastic approach appears to be the most accurate until one considers the importance of tracking an individual’s exposure history. For many pollutants, including carcinogens, establishing an exposure history is crucial, yet a probabilistic treatment of activity patterns makes this history impossible to determine. For example, a stochastic approach might indicate that 2.7% of the population are expected to be in the kitchen at 6:OO p.m. exposed to high concentrations of specific pollutants, and 13.9% in the living room at the same time. Similarly, this approach might indicate that at 7 : O O p.m. perhaps 1.2% of the population are expected to be in the kitchen and 16.8% in the dining room. However, the database does not permit us to specify the probability that an individual in a microenvironment during one time step is also there at another time step. The section on population activity data provides some detail on the pitfalls associated with determining even the simpler probability that some percentage of the population is in a specific microenvironment during a specific time period. At a minimum this kind of issue needs to be carefully evaluated before exposure model results are used in a risk assessment. Additional issues to consider in evaluating exposure models and their applicability to a given problem include the ability of the model to (1) treat profiles and dynamics applicable to the population group of interest, and ( 2 ) resolve geographic elements of interest. For example, models such as EXPOSURE may be applicable to a single building or building type for a single activity pattern, whereas models such as PAQM or SHAPE may be applicable to numerous building types and activity patterns. Environ. Sci. Tenhnol., Vol. 26, No. 5, 1992 853
I
BLE t
,Jmmary of representative indoor and exposure models
w* INDOOR (4)
wurce
Pollulant %Ink
Enter emission factor (mgW lor a heater) an0 act vitv level Wh’and hours of operation) lor each
Function of the surlace area 01 the walls h e r mav scec,tv a re-bmksio; rate
User specifies acn“ ceween chambers and outdoors A s sumes 100% mixing. Randomly alternates between natural and HVAC”-dominated aitflow. Variations in building configuration (open doors or windows), may be included in air flow parameters
User enters Not aooiicable PC-based, room vo,umes, Focus on moth HVAC ach, crystals. cigarettes. keroand ooerarina effictincv r%y sene heaters. for air cleaners and liwr wax. in HVAC sysModels one tem (when room or buildpresent) ina at a time. GGneralIv used lor residential buildinns
Minute. nowy, dail , or annuar average distributions 01 concentrations (pglm3) fo! , user-specified time period lor up to t 0 building types
User supplies User enters odtdoor con- reactivity faccenrratlons tors appropri(wnen aoolica- ale to wllurant he) and’hdoor bein ana emission rate lyzei. Sinks a (i.e., B/h) lor function of sureach “vera?-- .face ?!ea and ing period. n reactivity lacdoor sources tor. Re-emiscan be contin- sion rates uous or inter- must be incormittent. Models porated into one pollutant emission rates at a time by user
For eacn bu Iding ty e user specfe; mlxina factor. in11 trgtion, recirculated, and HVAC ach
For eacn bu. d. ing type. user enters HVAC characteristics (e.g., ach 01 outdoor and recirculated indoor air), air filter efficiencies, and indoor volume
Steady-state or dynamic concentration compute6 each roo modeled
User defines time.dependent rate for one or more species. Influx from outdoors specified as boundatv condition . .........*,.;: . . . ......
Roc... room and indoor1 outdoor iMerchanges. Onedimensional dispersive mixing addressed in ductwork. 100% mixing in each zone
Pollutant
Example output
pnmnlws
Distribution of hourly concentrations (pg! m3\ for each room modeled
Characlerlzation characterhation
?Za82’eY‘ one pollutant at a time
SI
4QI
‘-*
.... . . .
Multiple 1storder reactions considered. User deveioos own reiatioriship to malerials. exoosed areas, ‘chemistry, etc
. . ’. ..‘